The new molecular gastronomy, or, a gustatory tour of network analysis.

www.BioTechniques.com 19 Vol. 53 | No. 1 | 2012 The papaya platter was a mistake. On a hot afternoon in June, my family sat down to sample three dishes. Raspberry and kalamata olive tapenade with lemon zest and balsamic vinegar on crostini with prosciutto; chunks of papaya sprinkled with freshly grated Parmesan cheese and seared with a torch; and Oregon bleu cheese chocolate cake topped with a bleu cheese and cream cheese frosting. The tapenade was generally agreed to be delicious, the sweet balsamic vinegar complementing the zing of the fruit. But it was a tad too tart for my taste; I found the mixture of raspberry and citrus overpowering. The cake was phenomenal, with the bleu cheese in the frosting complementing the dark chocolate in the warmed cake to produce a delicious, if fleeting, bouquet. It was gone in no time. And then there was the parmesan papaya. That one was a dud, a goose egg. To me, the combination tasted vaguely of tuna, as in a Niçoise salad. We ate it, but we didn’t enjoy it. Mostly, we were trying to figure out exactly what was wrong with it. Stil l, two-out-of-three ain’t bad, especially considering these recipes weren’t pulled off the internet or the Food Network—they came from a systems biology network analysis study published in the journal Scientific Reports. A team of researchers—led by Albert-László Barabási, an expert on network analysis at Northeastern University and the Center for Cancer Systems Biology at the DanaFarber Cancer Institute who, among other things, builds protein-protein interaction networks with geneticist Marc Vidal— applied the computational tools of systems biology to analyze 56,498 recipes pulled from three online databases in an effort to test a hypothesis advanced by chef Heston Blumenthal, among others, “that ingredients sharing flavor compounds are more likely to taste well together than ingredients that do not” (1). For the study, the team first cross-referenced each recipe’s ingredients against Fenaroli’s Handbook of Flavor Ingredients to create a so-called bipartite network comprising nodes of both ingredients (such as shrimp, olive oil, or parsley) and the chemical compounds that give those ingredients flavor. From that, they constructed a “flavor network” projection in which ingredients, which are represented by nodes (dots), are connected by edges (lines) if the two ingredients have at least one flavor compound in common. 1-penten-3-ol, for instance, is shared by shrimp and parmesan cheese, while alpha-terpineol is found in parsley and tomato. The thickness of the lines indicates the number of compounds the two ingredients share, while the size of the nodes indicates how common the ingredient is in the analyzed recipes. The result is the sort of map found in interactome and genetic network studies, a multicolored star chart of sorts, except the stars in this case bear labels such as roasted hazelnut, macaroni, and tabasco pepper. Naturally, related foods tend to bunch up in this representation, with constellations for fish and fruits, spices and cheeses. Mushrooms cluster together, but are separate from the main network, as they share no flavor compounds with other ingredients. Interestingly, when Barabási’s team crossreferenced their network against the recipe database they found that North American and Western European recipes do in fact tend to pair ingredients with shared flavor compounds. But East Asian, specifically Korean, recipes do not. “To our surprise, we found that there seemed to be quite a large regional variation,” says Sebastian Ahnert, Royal Society University Research Fellow at the Cavendish Laboratory, one of the coauthors on the study.